Rework the compute expansions

[jthorton]

Currently to expand compute for a dataset a user has to make an empty openff.qcsubmit.dataset and just add the compute specifications they want and submit them again. This was originally a workaround which is not very flexible as it offers no subset support and only supports one optimization procedure per dataset. It might be time to adopt an official compute expansion pathway which is more flexible this would need to go into qcsubmit and it would be great to get some feedback on the proposal in this iusse.